On 27/06/2019 12:45, Chris Wilson wrote:
Quoting Lionel Landwerlin (2019-06-27 09:00:42)We want the ability to dispatch a set of command buffer to the hardware, each with a different OA configuration. To achieve this, we reuse a couple of fields from the execbuf2 struct (I CAN HAZ execbuf3?) to notify what OA configuration should be used for a batch buffer. This requires the process making the execbuf with this flag to also own the perf fd at the time of execbuf. v2: Add a emit_oa_config() vfunc in the intel_engine_cs (Chris) Move oa_config vma to active (Chris) v3: Don't drop the lock for engine lookup (Chris) Move OA config vma to active before writing the ringbuffer (Chris) v4: Reuse i915_user_extension_fn Serialize requests with OA config updates Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@xxxxxxxxx> --- .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 130 +++++++++++++++++- drivers/gpu/drm/i915/gt/intel_engine_cs.c | 2 + drivers/gpu/drm/i915/gt/intel_engine_types.h | 9 ++ drivers/gpu/drm/i915/gt/intel_lrc.c | 1 + drivers/gpu/drm/i915/gt/intel_ringbuffer.c | 4 +- drivers/gpu/drm/i915/i915_drv.c | 4 + drivers/gpu/drm/i915/i915_drv.h | 10 +- drivers/gpu/drm/i915/i915_perf.c | 26 ++-- include/uapi/drm/i915_drm.h | 37 +++++ 9 files changed, 208 insertions(+), 15 deletions(-) diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c index 476fade6fcab..18ae3e8a4e02 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c @@ -283,7 +283,11 @@ struct i915_execbuffer { struct { u64 flags; /** Available extensions parameters */ struct drm_i915_gem_execbuffer_ext_timeline_fences timeline_fences; + struct drm_i915_gem_execbuffer_ext_perf perf_config; } extensions; + + struct i915_oa_config *oa_config; /** HW configuration for OA, NULL is not needed. */ + struct drm_i915_gem_object *oa_bo; };#define exec_entry(EB, VMA) (&(EB)->exec[(VMA)->exec_flags - (EB)->flags])@@ -1210,6 +1214,34 @@ static int reloc_move_to_gpu(struct i915_request *rq, struct i915_vma *vma) return err; }++static int +get_execbuf_oa_config(struct drm_i915_private *dev_priv, + s32 perf_fd, u64 oa_config_id, + struct i915_oa_config **out_oa_config, + struct drm_i915_gem_object **out_oa_obj) +{ + struct file *perf_file; + int ret; + + if (!dev_priv->perf.oa.exclusive_stream) + return -EINVAL; + + perf_file = fget(perf_fd); + if (!perf_file) + return -EINVAL; + + if (perf_file->private_data != dev_priv->perf.oa.exclusive_stream) + return -EINVAL; + + fput(perf_file); + + ret = i915_perf_get_oa_config(dev_priv, oa_config_id, + out_oa_config, out_oa_obj); + + return ret; +} + static int __reloc_gpu_alloc(struct i915_execbuffer *eb, struct i915_vma *vma, unsigned int len) @@ -2072,6 +2104,64 @@ add_to_client(struct i915_request *rq, struct drm_file *file) list_add_tail(&rq->client_link, &rq->file_priv->mm.request_list); }+static int eb_oa_config(struct i915_execbuffer *eb)+{ + struct i915_vma *oa_vma; + int err; + + if (!eb->oa_config) + return 0; + + err = i915_active_request_set(&eb->engine->last_oa_config, + eb->request); + if (err) + return err; + + /* + * If the perf stream was opened with hold preemption, flag the + * request properly so that the priority of the request is bumped once + * it reaches the execlist ports. + */ + if (eb->i915->perf.oa.exclusive_stream->hold_preemption) + eb->request->has_perf = true;Leave this chunk for the next battle^W patch.
Oops.
+ /* + * If the config hasn't changed, skip reconfiguring the HW (this is + * subject to a delay we want to avoid has much as possible). + */ + if (eb->oa_config == eb->i915->perf.oa.exclusive_stream->oa_config) + return 0; + + oa_vma = i915_vma_instance(eb->oa_bo, + &eb->engine->i915->ggtt.vm, NULL); + if (unlikely(IS_ERR(oa_vma))) + return PTR_ERR(oa_vma); + + err = i915_vma_pin(oa_vma, 0, 0, PIN_GLOBAL); + if (err) + return err;Ugh. We should not be pinned after creating the request. Might not matter -- it's just reservation under its own lock that must not be crossed with the timeline lock currently held here.
Should I move this into get_execbuf_oa_config() ?
+ + err = i915_vma_move_to_active(oa_vma, eb->request, 0);You can unpin immediately after move_to_active (as the active is treated by being pinned). I'm thinking this will be replaced with just i915_active_ref() as you only care about the vma.+ if (err) { + i915_vma_unpin(oa_vma); + return err; + } + + err = eb->engine->emit_bb_start(eb->request, + oa_vma->node.start, + 0, I915_DISPATCH_SECURE); + if (err) { + i915_vma_unpin(oa_vma); + return err; + } + + i915_vma_unpin(oa_vma); + + swap(eb->oa_config, eb->i915->perf.oa.exclusive_stream->oa_config); + + return 0; +} + static int eb_submit(struct i915_execbuffer *eb) { int err; @@ -2098,6 +2188,10 @@ static int eb_submit(struct i915_execbuffer *eb) return err; }+ err = eb_oa_config(eb);+ if (err) + return err; + err = eb->engine->emit_bb_start(eb->request, eb->batch->node.start + eb->batch_start_offset, @@ -2520,8 +2614,22 @@ static int parse_timeline_fences(struct i915_user_extension __user *ext, void *d return 0; }+static int parse_perf_config(struct i915_user_extension __user *ext, void *data)+{ + struct i915_execbuffer *eb = data; + + if (copy_from_user(&eb->extensions.perf_config, ext, + sizeof(eb->extensions.perf_config))) + return -EFAULT; + + eb->extensions.flags |= BIT(DRM_I915_GEM_EXECBUFFER_EXT_PERF); + + return 0; +} + static const i915_user_extension_fn execbuf_extensions[] = { [DRM_I915_GEM_EXECBUFFER_EXT_TIMELINE_FENCES] = parse_timeline_fences, + [DRM_I915_GEM_EXECBUFFER_EXT_PERF] = parse_perf_config, };static int@@ -2573,6 +2681,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, eb.buffer_count = args->buffer_count; eb.batch_start_offset = args->batch_start_offset; eb.batch_len = args->batch_len; + eb.oa_config = NULL;eb.batch_flags = 0;if (args->flags & I915_EXEC_SECURE) { @@ -2651,9 +2760,23 @@ i915_gem_do_execbuffer(struct drm_device *dev, if (unlikely(err)) goto err_unlock;+ if (eb.extensions.flags & BIT(DRM_I915_GEM_EXECBUFFER_EXT_PERF)) {+ if (!intel_engine_has_oa(eb.engine)) { + err = -ENODEV; + goto err_engine; + } + + err = get_execbuf_oa_config(eb.i915, + eb.extensions.perf_config.perf_fd, + eb.extensions.perf_config.oa_config, + &eb.oa_config, &eb.oa_bo); + if (err) + goto err_engine;Why is this under the struct_mutex?
Access to dev_priv->perf.oa.exclusive_stream must be under struct_mutex. Also because we verify that the engine actually has OA support. I could split the getting the configuration part away.
+ } + err = eb_wait_for_ring(&eb); /* may temporarily drop struct_mutex */ if (unlikely(err)) - goto err_engine; + goto err_oa;err = eb_relocate(&eb);if (err) { @@ -2806,6 +2929,11 @@ i915_gem_do_execbuffer(struct drm_device *dev, err_vma: if (eb.exec) eb_release_vmas(&eb); +err_oa: + if (eb.oa_config) { + i915_gem_object_put(eb.oa_bo); + i915_oa_config_put(eb.oa_config); + } err_engine: eb_unpin_context(&eb); err_unlock: diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c index 4961f74fd902..ba4d2c1abd1a 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c @@ -859,6 +859,8 @@ int intel_engine_init_common(struct intel_engine_cs *engine)engine->set_default_submission(engine); + INIT_ACTIVE_REQUEST(&engine->last_oa_config);+ return 0;err_unpin:diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h index 7e056114344e..39da40937e7f 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_types.h +++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h @@ -363,6 +363,8 @@ struct intel_engine_cs { struct i915_wa_list wa_list; struct i915_wa_list whitelist;+ struct i915_active_request last_oa_config;+ u32 irq_keep_mask; /* always keep these interrupts */ u32 irq_enable_mask; /* bitmask to enable ring interrupt */ void (*irq_enable)(struct intel_engine_cs *engine); @@ -446,6 +448,7 @@ struct intel_engine_cs { #define I915_ENGINE_HAS_SEMAPHORES BIT(3) #define I915_ENGINE_NEEDS_BREADCRUMB_TASKLET BIT(4) #define I915_ENGINE_IS_VIRTUAL BIT(5) +#define I915_ENGINE_HAS_OA BIT(6) unsigned int flags;/*@@ -541,6 +544,12 @@ intel_engine_is_virtual(const struct intel_engine_cs *engine) return engine->flags & I915_ENGINE_IS_VIRTUAL; }+static inline bool+intel_engine_has_oa(const struct intel_engine_cs *engine) +{ + return engine->flags & I915_ENGINE_HAS_OA; +} + #define instdone_slice_mask(dev_priv__) \ (IS_GEN(dev_priv__, 7) ? \ 1 : RUNTIME_INFO(dev_priv__)->sseu.slice_mask) diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c index 28685ba91a2c..a6ea468986d1 100644 --- a/drivers/gpu/drm/i915/gt/intel_lrc.c +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c @@ -2763,6 +2763,7 @@ int intel_execlists_submission_setup(struct intel_engine_cs *engine) engine->init_context = gen8_init_rcs_context; engine->emit_flush = gen8_emit_flush_render; engine->emit_fini_breadcrumb = gen8_emit_fini_breadcrumb_rcs; + engine->flags |= I915_ENGINE_HAS_OA; }return 0;diff --git a/drivers/gpu/drm/i915/gt/intel_ringbuffer.c b/drivers/gpu/drm/i915/gt/intel_ringbuffer.c index f094406dcc56..a1aee092d0a3 100644 --- a/drivers/gpu/drm/i915/gt/intel_ringbuffer.c +++ b/drivers/gpu/drm/i915/gt/intel_ringbuffer.c @@ -2192,8 +2192,10 @@ static void setup_rcs(struct intel_engine_cs *engine) engine->irq_enable_mask = I915_USER_INTERRUPT; }- if (IS_HASWELL(i915))+ if (IS_HASWELL(i915)) { engine->emit_bb_start = hsw_emit_bb_start; + engine->flags |= I915_ENGINE_HAS_OA; + }engine->resume = rcs_resume;} diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c index 48f9009a6318..ef54cf1c54a5 100644 --- a/drivers/gpu/drm/i915/i915_drv.c +++ b/drivers/gpu/drm/i915/i915_drv.c @@ -486,6 +486,10 @@ static int i915_getparam_ioctl(struct drm_device *dev, void *data, case I915_PARAM_PERF_REVISION: value = 1; break; + case I915_PARAM_HAS_EXEC_PERF_CONFIG: + /* Obviously requires perf support. */ + value = dev_priv->perf.initialized; + break; default: DRM_DEBUG("Unknown parameter %d\n", param->param); return -EINVAL; diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 0323c8182a67..51b103493623 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -1096,6 +1096,8 @@ struct i915_oa_reg { };struct i915_oa_config {+ struct drm_i915_private *i915; + char uuid[UUID_STRING_LEN + 1]; int id;@@ -1114,7 +1116,7 @@ struct i915_oa_config { struct list_head vma_link; - atomic_t ref_count;+ struct kref ref; };struct i915_perf_stream;@@ -2612,6 +2614,12 @@ int i915_perf_get_oa_config(struct drm_i915_private *i915, int metrics_set, struct i915_oa_config **out_config, struct drm_i915_gem_object **out_obj); +void i915_oa_config_release(struct kref *ref); + +static inline void i915_oa_config_put(struct i915_oa_config *oa_config) +{ + kref_put(&oa_config->ref, i915_oa_config_release); +}/* i915_gem_evict.c */int __must_check i915_gem_evict_something(struct i915_address_space *vm, diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c index b2f5ba87921c..34d74fa64c74 100644 --- a/drivers/gpu/drm/i915/i915_perf.c +++ b/drivers/gpu/drm/i915/i915_perf.c @@ -366,10 +366,11 @@ struct perf_open_properties { int oa_period_exponent; };-static void put_oa_config(struct i915_oa_config *oa_config)+void i915_oa_config_release(struct kref *ref) { - if (!atomic_dec_and_test(&oa_config->ref_count)) - return; + struct i915_oa_config *oa_config = container_of(ref, typeof(*oa_config), ref); + + lockdep_assert_held(&oa_config->i915->perf.metrics_lock);That's not true for the eb paths.
Actually that's not true anywhere...
if (oa_config->obj) {list_del(&oa_config->vma_link); @@ -485,7 +486,7 @@ int i915_perf_get_oa_config(struct drm_i915_private *i915, }if (out_config) {- atomic_inc(&oa_config->ref_count); + kref_get(&oa_config->ref); *out_config = oa_config; }@@ -507,7 +508,7 @@ int i915_perf_get_oa_config(struct drm_i915_private *i915, err_buf_alloc:if (out_config) { - put_oa_config(oa_config); + i915_oa_config_put(oa_config); *out_config = NULL; } unlock: @@ -1483,7 +1484,7 @@ static void i915_oa_stream_destroy(struct i915_perf_stream *stream) if (stream->ctx) oa_put_render_ctx_id(stream);- put_oa_config(stream->oa_config);+ i915_oa_config_put(stream->oa_config);if (dev_priv->perf.oa.spurious_report_rs.missed) {DRM_NOTE("%d spurious OA report notices suppressed due to ratelimiting\n", @@ -2430,7 +2431,7 @@ static int i915_oa_stream_init(struct i915_perf_stream *stream, free_oa_buffer(dev_priv);err_oa_buf_alloc:- put_oa_config(stream->oa_config); + i915_oa_config_put(stream->oa_config);intel_uncore_forcewake_put(&dev_priv->uncore, FORCEWAKE_ALL);intel_runtime_pm_put(&dev_priv->runtime_pm, stream->wakeref); @@ -3246,7 +3247,7 @@ void i915_perf_register(struct drm_i915_private *dev_priv) if (ret) goto sysfs_error;- atomic_set(&dev_priv->perf.oa.test_config.ref_count, 1);+ dev_priv->perf.oa.test_config.i915 = dev_priv;goto exit; @@ -3502,7 +3503,8 @@ int i915_perf_add_config_ioctl(struct drm_device *dev, void *data,return -ENOMEM; }- atomic_set(&oa_config->ref_count, 1);+ oa_config->i915 = dev_priv; + kref_init(&oa_config->ref);if (!uuid_is_valid(args->uuid)) {DRM_DEBUG("Invalid uuid format for OA config\n"); @@ -3601,7 +3603,7 @@ int i915_perf_add_config_ioctl(struct drm_device *dev, void *data, sysfs_err: mutex_unlock(&dev_priv->perf.metrics_lock); reg_err: - put_oa_config(oa_config); + i915_oa_config_put(oa_config); DRM_DEBUG("Failed to add new OA config\n"); return err; } @@ -3655,7 +3657,7 @@ int i915_perf_remove_config_ioctl(struct drm_device *dev, void *data,DRM_DEBUG("Removed config %s id=%i\n", oa_config->uuid, oa_config->id); - put_oa_config(oa_config);+ i915_oa_config_put(oa_config);config_err:mutex_unlock(&dev_priv->perf.metrics_lock); @@ -3824,7 +3826,7 @@ static int destroy_config(int id, void *p, void *data) { struct i915_oa_config *oa_config = p;- put_oa_config(oa_config);+ i915_oa_config_put(oa_config);return 0;} diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index b7fe1915e34d..5938a73963fe 100644 --- a/include/uapi/drm/i915_drm.h +++ b/include/uapi/drm/i915_drm.h @@ -623,6 +623,16 @@ typedef struct drm_i915_irq_wait { */ #define I915_PARAM_HAS_EXEC_TIMELINE_FENCES 55+/*+ * Request an i915/perf performance configuration change before running the + * commands given in an execbuf. + * + * Performance configuration ID and the file descriptor of the i915 perf + * stream are given through drm_i915_gem_execbuffer_ext_perf. See + * I915_EXEC_EXT. + */ +#define I915_PARAM_HAS_EXEC_PERF_CONFIG 56 + /* Must be kept compact -- no holes and well documented */typedef struct drm_i915_getparam {@@ -1026,6 +1036,12 @@ enum drm_i915_gem_execbuffer_ext { */ DRM_I915_GEM_EXECBUFFER_EXT_TIMELINE_FENCES,+ /**+ * This identifier is associated with + * drm_i915_gem_execbuffer_perf_ext. + */ + DRM_I915_GEM_EXECBUFFER_EXT_PERF, + DRM_I915_GEM_EXECBUFFER_EXT_MAX /* non-ABI */ };@@ -1054,6 +1070,27 @@ struct drm_i915_gem_execbuffer_ext_timeline_fences {__u64 values_ptr; };+struct drm_i915_gem_execbuffer_ext_perf {+ struct i915_user_extension base; + + /** + * Performance file descriptor returned by DRM_IOCTL_I915_PERF_OPEN. + * This is used to identify that the application + */ + __s32 perf_fd; + + /** + * Unused for now. Must be cleared to zero. + */ + __u32 pad; + + /** + * OA configuration ID to switch to before executing the commands + * associated to the execbuf. + */ + __u64 oa_config; +}; + struct drm_i915_gem_execbuffer2 { /** * List of gem_exec_object2 structs -- 2.21.0.392.gf8f6787159e _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
